The present invention relates to a storage system, particularly relates to control for dispersing a load in a storage system having cluster configuration and the control of a cache memory for it.
In an enterprise IT system, a mass and high-performance storage system is demanded.
To meet this demand, mass data can be processed by adopting plural storage systems of small storage capacity. However, as the number of storage systems increases, the increase of the management cost of the storage systems by failure, maintenance and others comes into question. In the meantime, there is also a method of providing a mass storage system by one storage system. However, in a conventional type storage system that a computer resource such as a memory, a control memory and an internal transfer mechanism is shared, and it is difficult to realize a currently demanded mass and high-performance storage system because of the cost and a technical reason.
To solve such a problem, in a specification of U.S. Pat. No. 6,256,740 for example, the application of cluster technique to a storage system is disclosed. The cluster technique has been mainly used for packaging technique for realizing great throughput in a field of a host computer such as a server. A large-scale storage system can be mounted at a relatively low cost by applying this technique to the storage system. Such a storage system is called a cluster storage system.
As for the cluster storage system, plural storage nodes of relatively small configuration are connected via an internal network and a mass single storage system is realized. An I/O request for the cluster storage system is portioned out between (among) storage nodes that hold target data and is processed in each storage node. At the storage node, a host interface, a disk device, a control processor, a memory, a control memory, a disk cache and others are mounted like a normal storage system and these are connected via the internal network at the storage node. At each storage node, an I/O request for the disk device is made using these.
Generally, when a large-scale cache memory and a large-scale control memory are managed as shared memory space, a broad-band internal network is required to correspond to mass access and the cost of a storage system increases. However, as the scale of a storage node is small, a load of access to a cache memory and a control memory is small and the cost of a storage system can be reduced. Therefore, a cluster storage system can realize a mass storage system at a low cost.
Besides, in JP-A-10-283272, there is disclosed a compound computing system in which data generated in an I/O subsystem of an open system is transferred to an I/O subsystem of a main frame to which the subsystem is not directly connected between the I/O subsystems different in an access interface and is backed up in a storage of the I/O subsystem of the main frame. A disk controller of the main frame is provided with a table showing whether or not an address of its storage is allocated to the I/O subsystem of the open system and allows access to the I/O subsystem of the open system by referring to this table.